agent-terminal-ui¶
agent-terminal-ui is a Textual-based terminal
user interface for interacting with AI agents built on the
agent-utilities platform. It connects
to an agent backend over dual protocols — AG-UI (SSE streaming) and ACP
(JSON-RPC + SSE) — and renders live workflow activity, tool execution, and
human-in-the-loop approvals in the terminal.
This site is the official documentation for the client: its architecture, configuration surface, durable session infrastructure, and the autonomous goal and multi-session workflows it supports.
Highlights¶
- Lightweight by design — the client is a thin frontend (~60–85 MB interactive,
~30 MB headless). It never imports the heavy
agent_utilitiesbackend; all weight stays in one shared backend service reached over HTTP/SSE. - Interactive or headless — run the full TUI, or
agent-terminal-ui --headless --prompt "…"for many concurrent non-interactive sessions against one backend. - Live workflow & tools — a dynamic workflow sidebar, streaming Markdown responses, expandable tool-call blocks, and human-in-the-loop approvals.
- Durable & recoverable — SQLite-backed sessions, pre-turn checkpoints, and
side-git workspace snapshots with
/restore N.
Documentation map¶
- Architecture — protocol connectivity, key components, environment variables, and implementation details.
- Features — slash commands, keyboard shortcuts, and the model picker.
- Configuration — the complete settings reference, hooks configuration, and sandbox modes.
- Session Management — durable session persistence, pre-turn checkpointing, crash recovery, and workspace snapshots.
- Agent View — the multi-session dashboard and background agent management.
- Goal Command — the autonomous
/goalloop and its Knowledge-Graph-native integration. - Agents & Issues — known issues, recent changes, and the session journal.
- Concepts — the stable concept registry (
CONCEPT:TUI-*) tracing the client's core ideas across documentation and source.
Installation¶
The client is published on PyPI:
Launch the terminal interface:
Consult the Configuration reference to point the client at your agent backend and tune its behavior.